Concepts in Machine Learning for Physicists

Learning Outcomes

Subject Specific Intellectual and Research Skills

Having successfully completed this module you will be able to:

• interpret data using statistical techniques and make decisions taking into account experimental errors
• apply machine learning techniques and artificial intelligence to a wide variety of physical problems
• program and use computers to assist in the solution of physical problems

Knowledge and Understanding

Having successfully completed this module, you will be able to demonstrate knowledge and understanding of:

• mathematics required for the description of the physical world
• principles and implementation of machine learning techniques and artificial intelligence
• general principles of coding in Python and other languages and applications in Physics

1. Introduction to Machine Learning. Historical context, biological origins. Development and current status. Overview of ML & AI applications. 2. Mathematical Objects: Variables and functions. Functions of single and multiple variables. Data and representation of data: arrays, matrices, tensors. Functions as operators (matrices). Connections with linear algebra. Algorithms. Concepts will be elaborated using python. 3. Review of probability and statistics. Probability distributions: discrete and continuous. Data (sets) used in physics and astronomy. Statistical treatment of data. Error analysis. 4. Models to describe data. Estimation of means and errors of parameters in a model. Confidence intervals. Regression and hypothesis testing. (Review from PHYS1201). 5. Modelling as a machine learning exercise: supervised, unsupervised, reinforcement learning, Generative AI. Qualitative descriptions of each category with examples. 6. Linear Regression from ML. Data, Model, Training, Evaluating, Inference. Physics-oriented example illustrating the five steps. Concepts of measuring loss, gradient descent to optimize the loss, and hyperparameter tuning will be introduced. 7. Logistic regression as a binary classifier. The sigmoid function. Measures of loss, optimization and regularization. Computation of probability from output and prediction.

The aim of this module is to give students practical skills, so the teaching and learning methods used are designed to accomplish this. Formal lectures will be used primarily to introduce key ideas and concepts, but even these will be illustrated with extensive practical/computational examples and visualisations. Much of the teaching will take place during extended "practical" sessions, during which students will be expected to carry out programming tasks that are related to -- and illustrative of -- the concepts that are being explored in the module at that time. Ideally, the formal lecture content will take place immediately before or during these sessions, so that new theoretical concepts being introduced can immediately be explored in practice by students. Teaching support in the form of multiple demonstrators will be available during all sessions, so that one-on-one help is available as needed. Additional learning is expected to take place independently, again mostly in the form of practical programming.

Summative

This is how we’ll formally assess what you have learned in this module.

Referral

This is how we’ll assess you if you don’t meet the criteria to pass this module.